Consonant Gradation in Estonian and Sámi: Two-Level Solution

Koskenniemi’s two-level morphology was the first practical general model in the history of computational linguistics for the analysis of morphologically complex languages. In this article we will reconsider one of the key innovations in Koskenniemi (1983), namely the treatment of consonant gradation in finite state transducers. We will look not at Finnish, but at two languages with a more extensive consonant gradation system, namely Estonian and Sámi. The goal of the paper is to demonstrate two different ways of modeling consonant gradation in a finite state morphological system lexical and morphophonological. We will also compare the resulting systems by their computational complexity and human-readability. Consonant gradation is rare among the languages of the world, but stem alternation in itself is not, and the treatment of consonant gradation can readily be transferred to other stem alternation phenomena. Koskenniemi’s original idea was to see stem alternation as an agglutinative phenomenon. Consider the example (14.1), showing a two-level representation of stem alternation.